Systems and Methods for Facial Property Identification

• US 20140341422A1
• Filed: 01/08/2014
• Published: 11/20/2014
• Est. Priority Date: 05/10/2013
• Status: Active Grant

• Current Assignee: Tencent Technology Company Limited (Tencent Holdings Limited)
• Original Assignee: Tencent Technology Company Limited (Tencent Holdings Limited)
• Inventors: Xiong, Pengfei, Liu, Hailong, Chen, Bo, Liu, Yan
• Granted Patent: US 9,679,195 B2
• Time in Patent Office: Days
• US Class Current: 382/103
• CPC Class Codes: G06F 18/2413 based on distances to training or reference patterns
• CPC Class Codes: G06V 40/169 Holistic features and representations, i.e. based on the facial image taken as a whole
• CPC Class Codes: G06V 40/172 Classification, e.g. identification
• CPC Class Codes: G06V 40/178 estimating age from face image; using age information for improving recognition

Abstract

Systems and methods are provided for facial property identification. For example, an image sample is acquired; a first effective area image of a face is acquired in the image sample; first textural features of the first effective area image are extracted: and the first textural features of the first effective area image are classified by race, gender and age using a race classifier, a gender classifier and an age classifier successively to obtain a race property, a gender property and an age property of the face.

Claims

1. A method for facial property identification, the method comprising:

acquiring an image sample;

acquiring a first effective area image of a face in the image sample;

extracting first textural features of the first effective area image; and

classifying the first textural features of the first effective area image by race, gender and age using a race classifier, a gender classifier and an age classifier successively to obtain a race property, a gender property and an age property of the face.

2. The method of claim 1, further comprising:

establishing a race classifier, a gender classifiers and an age classifier successively.

3. The method of claim 2, wherein the establishing a race classifier, a gender classifiers and an age classifier successively comprises:

acquiring a training image sample collection;

acquiring a second effective area image in an image sample in the training image sample collection;

extracting second textural features of the second effective area image; and

establishing the race classifier, the gender classifier and the age classifier successively based on at least information associated with the second textural features of the second effective area images in the training image sample collection.

4. The method of claim 3, wherein the establishing the race classifier based on at least information associated with the second textual features in the training image sample collection comprises:

dividing the training sample collection into white training samples, black training samples and yellow training samples corresponding to a white race, a black race and a yellow race respectively; and

training third textural features of the white training samples, the black training samples and the yellow training samples separately to obtain a ternary classifier associated with the white race, the black race and the yellow race.

5. The method of claim 4, wherein the establishing the gender classifier based on at least information associated with the second textural features in the training image sample collection comprises:

dividing the white training samples, the black training samples and the yellow training samples by gender to obtain gender samples of each race; and

training fourth textural features of the gender samples of each race separately to obtain a gender binary classifier associated with each race, wherein the binary classifier corresponds to male and female.

6. The method of claim 5, wherein the establishing the age classifier based on at least information associated with the second textural features in the training image sample collection comprises:

dividing the gender samples of each race by age groups to obtain infant training samples, child training samples, youth training samples and senior training samples;

training the infant training samples, the child training samples, the youth training samples and the senior training samples to establish a first-level age classifier;

training fifth textural features of second-level training samples associated with an age interval of five years related to the infant training samples, the child training samples, the youth training samples and the senior training samples to establish a second-level age classifier; and

training a linearly-fit third-level age classifier for the infant training samples, the child training samples, the youth training samples and the senior training samples based on at least information associated with an age interval of five years.

7. The method of claim 1, wherein the acquiring a first effective area image of a face in the image sample includes:

detecting the face in the image sample;

determining eye positions on the face;

calibrating an original image of the face based on the eye positions on the face; and

obtaining the first effective area image of the face within a predetermined area centering on the eyes of the face.

8. The method of claim 1, wherein the extracting first textural features of the first effective area image includes:

extracting biologically inspired model (BIM) features within the first effective area image of the face.

9. The method of claim 8, wherein the extracting BIM features within the first effective area image of the face comprises:

establishing step, establishing 64 sets of gabor filters and filtering the first effective area image of the face in 16 scales and 4 orientations using the 64 sets of gabor filters to obtain a gabor image;

dividing step, dividing the gabor image into 8 parts, wherein a part includes two scales and four orientations;

selecting step, in an orientation of a part, selecting one set of m×

n masks and dividing the gabor image to obtain two sets of serial gabor features; and

comparison step, comparing gabor features corresponding to the two scales of the part and taking a higher value of corresponding feature dimensions as a final feature output;

adjusting step, adjusting sizes of the masks; and

repeating the operations the selecting step to the comparison step for k times to obtain k×

4×

8 sets of BIM features of the first effective area image of the face.

10. The method of claim 9, further comprising:

performing principal component analysis (PCA) dimensionality reduction on the k×

4×

8 sets of BIM features to obtain the first textual features.

11. The method of claim 1, wherein the classifying the first textural features of the first effective area image by race, gender and age using a race classifier, a gender classifier and an age classifier successively to obtain a race property, a gender property and an age property of the face comprises:

classifying the first textural features of the first effective area image by race with a race classifier to obtain a race property of the face;

selecting a gender classifier corresponding to the race property of the face;

classifying the first textual features of the first effective area image by gender with the gender classifier to obtain a gender property of the face;

selecting an age classifier corresponding to the race property and the gender property of the face; and

classifying the first textural features of the first effective area image by age with the age classifier to obtain an age property of the face.

12. The method of claim 11, wherein the selecting a gender classifier corresponding to the race property of the face and the classifying the first textual features of the first effective area image by gender with the gender classifier to obtain a gender property of the face comprises:

selecting a first-level age classifier corresponding to the race property and the gender property of the face;

inputting the first textural features of the first effective area image into the first-level age classifier to obtain first-level age groups corresponding to the first effective area image and first weights of the first-level age groups;

selecting a second-level age classifier according to a particular first-level age group with a highest first weight among the first-level age groups;

inputting the first textural features of the first effective area image into the second-level age classifier to obtain second-level age groups corresponding to the first effective area image and second weights of the second-level age groups;

selecting a third-level age classifier according to a particular second-level age group with a highest second weight among the second-level age groups;

inputting the first textural features of the first effective area image into the third-level age classifier to obtain a third-level age corresponding to the first effective area image; and

obtaining an age property of the face according to the third-level age, the first weights of the first-level age groups and the second weights of the second-level age groups.

13. A system for facial property identification, comprising:

a test-sample-acquisition module configured to acquiring an image sample;

an effective-area-image-acquisition module configured to acquire a first effective area image of a face in the image sample;

a textural-feature-extraction module configured to extract first textural features of the first effective area image; and

a facial-property-identification module configured to classify the first textural features of the first effective area image by race, gender and age using a race classifier, a gender classifier and an age classifier successively to obtain a race property, a gender property and an age property of the face.

14. The system of claim 13, further comprising:

a property-classifier-training module configured to establish a race classifier, a gender classifiers and an age classifier successively.

15. The system of claim 14, wherein the property-classifier-training module includes:

a training-image-sample-collection-acquisition unit configured to acquire a training image sample collection;

an effective-area-image-acquisition unit configured to acquire a second effective area image in an image sample in the training image sample collection;

a textural-feature-extraction unit configured to extract second textural features of the second effective area image; and

a property-classifier-training unit configured to establish the race classifier, the gender classifier and the age classifier successively based on at least information associated with the second textural features of the second effective area images in the training image sample collection.

16. The system of claim 15, wherein the property-classifier-training unit comprises:

a race-classifier-training subunit configured to divide the training sample collection into white training samples, black training samples and yellow training samples corresponding to a white race, a black race and a yellow race respectively; and

train third textural features of the white training samples, the black training samples and the yellow training samples separately to obtain a ternary classifier associated with the white race, the black race and the yellow race.

17. The system of claim 16, wherein the property-classifier-training unit comprises:

a gender-classifier-training subunit configured to divide the white training samples, the black training samples and the yellow training samples by gender to obtain gender samples of each race and train fourth textural features of the gender samples of each race separately to obtain a gender binary classifier associated with each race, wherein the binary classifier corresponds to male and female.

18. The system of claim 17, wherein the property-classifier-training unit comprises:

an age-classifier-training subunit configured to;

divide the gender samples of each race by age groups to obtain infant training samples, child training samples, youth training samples and senior training samples;

train the infant training samples, the child training samples, the youth training samples and the senior training samples to establish a first-level age classifier;

train fifth textural features of second-level training samples associated with an age interval of five years related to the infant training samples, the child training samples, the youth training samples and the senior training samples to establish a second-level age classifier; and

train a linearly-fit third-level age classifier for the infant training samples, the child training samples, the youth training samples and the senior training samples based on at least information associated with an age interval of five years.

19. The system of claim 13, wherein the effective-area-image-acquisition module comprises:

a positioning unit configured to detect the face in the image sample and determine eye positions on the face; and

an effective-area-image-cutting unit configured to calibrate an original image of the face based on the eye positions on the face and obtain the first effective area image of the face within a predetermined area centering on the eyes of the face.

20. The system of claim 13, wherein the textural-feature-extraction module comprises:

a BIM-feature-extraction unit configured to extract biologically inspired model (BIM) features within the first effective area image of the face.

21. The system of claim 20, wherein the BIM-feature-extraction unit is further configured to:

a) establish 64 sets of gabor filters and filtering the first effective area image of the face in 16 scales and 4 orientations using the 64 sets of gabor filters to obtain a gabor image;

b) divide the gabor image into 8 parts, wherein a part includes two scales and four orientations;

c) in an orientation of a part, select one set of m×

n masks and dividing the gabor image to obtain two sets of serial gabor features; and

d) compare gabor features corresponding to the two scales of the part and taking a higher value of corresponding feature dimensions as a final feature output;

adjust sizes of the masks; and

repeat the operations c)-d) for k times to obtain k×

4×

8 sets of BIM features of the first effective area image of the face.

22. The system of claim 21, wherein the BIM-feature-extraction unit is further configured to perform principal component analysis (PCA) dimensionality reduction on the k×

4×

8 sets of BIM features to obtain the first textual features.

23. The system of claim 13, wherein the facial-property-identification module comprises:

a race-property-identification unit configured to classify the first textural features of the first effective area image by race with a race classifier to obtain a race property of the face;

a gender-property-identification unit configured to select a gender classifier corresponding to the race property of the face and classify the first textual features of the first effective area image by gender with the gender classifier to obtain a gender property of the face; and

an age-property-identification unit configured to select an age classifier corresponding to the race property and the gender property of the face and classify the first textural features of the first effective area image by age with the age classifier to obtain an age property of the face.

24. The system of claim 23, wherein the age-property-identification unit is further configured to:

select a first-level age classifier corresponding to the race property and the gender property of the face;

input the first textural features of the first effective area image into the first-level age classifier to obtain first-level age groups corresponding to the first effective area image and first weights of the first-level age groups;

select a second-level age classifier according to a particular first-level age group with a highest first weight among the first-level age groups;

input the first textural features of the first effective area image into the second-level age classifier to obtain second-level age groups corresponding to the first effective area image and second weights of the second-level age groups;

select a third-level age classifier according to a particular second-level age group with a highest second weight among the second-level age groups;

input the first textural features of the first effective area image into the third-level age classifier to obtain a third-level age corresponding to the first effective area image; and

obtain an age property of the face according to the third-level age, the first weights of the first-level age groups and the second weights of the second-level age groups.

25. The system of claim 13, further comprising:

one or more data processors; and

a computer-readable storage medium;

wherein one or more of the test-sample-acquisition module, the effective-area-image-acquisition module, the textural-feature-extraction module, and the facial-property-identification module are stored in the storage medium and configured to be executed by the one or more data processors.

26. A non-transitory computer readable storage medium comprising programming instructions for facial property identification, the programming instructions configured to cause one or more data processors to execute operations comprising:

acquiring an image sample;

acquiring a first effective area image of a face in the image sample;

extracting first textural features of the first effective area image; and

classifying the first textural features of the first effective area image by race, gender and age using a race classifier, a gender classifier and an age classifier successively to obtain a race property, a gender property and an age property of the face.